Brain Tumor Detection Using Transfer Learning
To frame automated segmentation and classification of brain tumors, around 3000 MRI images (both tumors and non-tumors) are collected and Otsu’s segmentation algorithm accuracy is obtained before and after the segmentation using four classification algorithms.
Abstract
The main objective of the proposed work is to encounter the most serious condition of brain tumors. However, if caught early enough, a brain tumor can be cured. MRI scans and CT scans are used to diagnose brain tumors in most cases. It is far too difficult to accurately detect a tumor’s location and size. It is often difficult for doctors and patients to comprehend the outcomes. This paper targets to frame automated segmentation and classification of brain tumors. In this work, around 3000 MRI images (both tumors and non-tumors) are collected. To identify the images with tumors, Otsu's segmentation method is used. Following this process, the feature extraction technique is carried out using the Vantage Point Tree algorithm, which employs the modification features of the VP Tree algorithm (grayscale images and removal of duplicate images) to increase accuracy, but the present VP Tree algorithm is used for color images. Next, Otsu’s segmentation algorithm accuracy is obtained before and after the segmentation using four classification algorithms namely Efficient Net B0, Efficient Net V2-B0, Res Net 50, Res Net 101, and VGG 19. Following this severity classification is carried out to categorize the datasets depending on their severity (i.e., Grade I, Grade II, Grade III, and Grade IV). Comparing the obtained results, the EfficientNetB0 and EfficientNetV2-B0 outperform in terms of accuracy, precision, and F1 score compared to the other classification algorithms.